Built an information retrieval model to pick the most relevant answers from the US tax law book for the input
question. Read and implemented the research paper “A Latent Semantic Model with Convolutional-Pooling
structure for Information Retrieval” from scratch in Pytorch.

Used pre trained CRAFT model to detect text regions in images and used tesseract to extract text from the regions.
Also, implemented the EAST algorithm from scratch in pytorch to detect text regions in images

Built a classification model to predict the attrition probability of each employee for the future financial year based on
historic data. Built an attrition prediction product in dash and deployed it on Azure web app to be used by the clients

Built a sentiment analysis model to analyze the live tweets about the organization. Converted the predicted
sentiments into a quantifiable score. Led a team of 4 through the entire lifecycle of the project along with handling
client conversations.

Performed time series clustering analysis to group stores based on their sales pattern and used these clusters to
forecast weekly sales for the stores, thus reducing the complexity involved in the store-level forecast process.

Built an information retrieval model to pick the most relevant answers from the US tax law book for the inputquestion. Read and implemented the research paper A Latent Semantic Model with Convolutional-Poolingstructure for Information Retrieval from scratch in Pytorch.

Built a classification model to predict the attrition probability of each employee for the future financial year based onhistoric data. Built an attrition prediction product in dash and deployed it on Azure web app to be used by the clients

Used pre trained CRAFT model to detect text regions in images and used tesseract to extract text from the regions.Also, implemented the EAST algorithm from scratch in pytorch to detect text regions in images.

Built an image classification model by experimenting over numerous CNN architectures, optimization algorithmsand image augmentations. Used transfer learning (Resnet18 and Resnet34) and re-trained the model.

Built a sentiment analysis model to analyze the live tweets about the organization. Converted the predictedsentiments into a quantifiable score. Led a team of 4 through the entire lifecycle of the project along with handlingclient conversations.

Performed time series clustering analysis to group stores based on their sales pattern and used these clusters toforecast weekly sales for the stores, thus reducing the complexity involved in the store-level forecast process.

Used pytorch, keras and transfer learning to build the model

Implement Siamese net LSTM architecture on crowdflower search result relevance competition dataset (Kaggle)

Implement LSTM model to predict entities on conll2003 dataset